AI Token Prices Halve, But Consumption Boom Drives Higher Spending, UBS Says

The cost of using artificial intelligence services has fallen sharply, yet this is fueling a surge in consumption that is leading to higher overall spending on the technology. This dynamic creates a positive feedback loop that could sustain elevated investment in AI infrastructure, according to a new analysis from UBS.

The Jevons Paradox in AI

The phenomenon reflects a classic economic principle known as the Jevons paradox, where greater efficiency in resource use lowers costs and, in turn, spurs an increase in overall consumption. According to UBS, the cost of AI tokens—the basic units of data processed by large language models (LLMs)—has dropped significantly.

• The LLM Token Expenditure Index fell from approximately $2 per million tokens in May to around $1 in August.
• This price drop has not curbed spending but has instead accelerated demand for AI computing power.

Surging Consumption and Spending

The trend is clearly visible on platforms like OpenRouter, which provides access to hundreds of different LLMs. The platform has seen a dramatic rise in both usage and high-value spending, underscoring the market's expanding appetite.

Data from OpenRouter cited by UBS shows that since the beginning of the year:

Ad

• Token usage on the platform has increased tenfold.
• Average monthly spending among its top 1% of users has climbed from $2,500 to $7,500.

This explosion in token consumption strengthens the business case for accelerated investment in the underlying infrastructure, including data centers and high-performance computing equipment.

The 'AI Flywheel' Effect

UBS describes this cycle as an “AI flywheel”: cheaper AI services enable greater usage, which supports demand for more computing capacity, and in turn encourages further investment and innovation. This momentum is amplified by significant efficiency gains in both hardware and the AI models themselves.

A study by The Economist, cited by UBS, found that the token cost to achieve specific performance benchmarks had plummeted by about 97% year-over-year. This continuous improvement in efficiency feeds the flywheel, driving broader and more intensive adoption.

For investors, UBS notes that token pricing and consumption are becoming critical indicators of activity across the AI economy. Rather than signaling weakness, falling token prices are evidence that AI adoption is broadening rapidly, pushing aggregate consumption and total spending higher.